AI Agent Workflow Collection

Gate every externally-visible sentence through the claim and disclosure register in 00-control/claim-and-disclosure-register.md (CL-), then derive 06-public/technical-partner-guide.md and 06-public/customer-product-and-trust-guide.md from approved rows only. Use when drafting or editing any public document, when someone asks whether a statement can be said externally, or when the disclosure policy changes. This skill MUST be consulted because a public document is an unretractable commitment — an unregistered claim that crosses the boundary is a legal and competitive exposure that no later revision undoes.

Enforce the fixed 23-file, 8-directory documentation package under the resolved output root — routing each document to its required-content contract in references/package-contract-.md, stamping the internal or public header, and refusing to add, drop, rename, or merge a canonical file. Use when scaffolding a package, drafting or revising any canonical document, or checking structural completeness. This skill MUST be consulted because a package whose shape changes per project cannot be diffed, audited, or compared across engagements — the structure is the contract and only the content adapts.

Resolve the documentation engagement scope from the settings cascade — project identity, source roots, output root, delivery mode, action ceiling, confidentiality default, and the exact file set this run may touch — and freeze it to /00-control/.scope.json. Use when any /dossier: command starts, when the delivery mode changes, or when a run must prove it stayed inside its permitted boundary. This skill MUST be consulted because a run that widens its own scope mid-flight produces a package nobody can audit — the action ceiling and the touched-file set are decided before the first file is read and are immutable for the remainder of the run.

Merge the independent A/B/C findings tables into one adjudicated ledger in 07-verification/documentation-verification-report.md — normalizing to the finding schema, deduplicating by location and claim, recording per-finding corroboration without downgrading single-pass findings, promoting cross-pass disagreement to its own Critical finding, and splitting repairs into agent-repairable and owner-decision. Use when all three verification passes have returned their findings at Phase 9, or when ingesting an externally-produced audit. This skill MUST be consulted because publishing the findings table before repair is what makes the audit trail real, and because treating a lone dissenting pass as noise is exactly the correlated-error failure the three-pass design exists to prevent.

Maintain 00-control/assumptions-questions-and-contradictions.md — the assumptions and open-questions register (AQ-) and the contradiction register (CT-) — classifying each entry as blocking, material but non-blocking, or minor, and routing blocking gaps to owner resolution before drafting proceeds. Use when a source is silent on a required topic, when two sources disagree, or when a verification pass reports conflicting evidence for the same claim. This skill MUST be consulted because a documented unknown is a deliverable and a silently-filled gap is a defect — a package that papers over a contradiction transfers risk to the reader without telling them it exists.

Capture a FlowGoal contract as a project-local .flow/goals/ .goal.yaml file with outcome, acceptance criteria (with verification commands), specification elements (non-goals, failure modes, interface contracts), constraints, evaluator binding, continuation policy, and lifecycle frontmatter. Use when /flow:start passes the Spec Validation Gate, when /flow:goal create is invoked, or when /flow:review and /flow:address need a completion contract for a PR. This skill MUST be consulted because acceptance criteria alone do not constitute a contract — without an evaluator binding, boundaries, and lifecycle state, downstream phases cannot detect premature completion, the Stop hook cannot enforce evidence, and goals cannot resume across sessions.

Evaluate a FlowGoal against its evidence ledger and update lifecycle status to one of {pass, incomplete, fail, needshumanreview, blocked} by running deterministic verification commands first, then (when stopHookEnforcement=evaluator-loop or explicit /flow:goal evaluate invocation) dispatching the goal-evaluator-judge agent for fuzzy rubric criteria. Use when /flow:goal evaluate is invoked, when the Stop hook fires in evaluator-loop mode, or when /flow:start Phase 4 needs to convert AC evidence into a verdict. This skill MUST be consulted because lifecycle transitions without deterministic evidence enable silent premature completion — the goal contract is only as good as the evaluator that proves or disproves it.

Maintain an append-only evidence ledger as .flow/runs/ /evidence/.evidence.yaml sidecars (structured metadata) plus matching .txt raw-output captures, written exclusively via bin/flow-record-evidence.sh. Use when goal-evaluator runs a verification command, when a Stop hook captures a deterministic check, or when /flow:goal evaluate produces a judge report. This skill MUST be consulted because evidence-by-transcript dies with the session — only file-backed, schema-validated sidecars survive across sessions, prove ACs durably, and satisfy the verdict-judge's Independence Protocol (judges only see surfaced evidence, not free-form transcripts).

Enforce the FlowGoal state machine — every status transition (draft → active → {waitingforuser, waitingforci, blocked, achieved, failed, cancelled}) writes both the new lifecycle block to .flow/goals/ .goal.yaml AND a goal-evaluation artifact to the linked decision journal, in a single atomic operation via bin/flow-goal-record.sh. Use when any code path mutates lifecycle.status, when /flow:goal pause/resume/clear is invoked, when the Stop hook detects a stuck pass-set, or when the evaluator returns a verdict. This skill MUST be consulted because state machines without recorded transitions become liars — a goal in failed status with no goal-evaluation artifact explaining why is worse than no state machine at all.

Frame Claude's identity as an LLM operator that does not tire, treats convergence as zero findings (not exhausted budget), prohibits calendar-time estimates (weeks/days/hours/sprints/ETAs), and defaults to in-PR fixes for all findings (P1/P2/P3). Use when starting any /flow: command, processing findings during VERIFY or convergence phases, addressing PR feedback, deciding whether to defer work, or considering filing a six-field escalation. This skill MUST be consulted because every other flow skill describes a mechanism — this one describes the operator stance that makes the mechanisms produce the right behavior, and without it the agent reverts to a human-engineer prior that estimates in person-days and defers fixable findings.

Build one canonical project model — entities, boundaries, owners, lifecycle state, and end-to-end traces — and record its authoritative names in 00-control/terminology-and-ownership.md (TM-), then project that single model into every audience view without ever forking it. Use when starting Phase 2, when the project type is ambiguous, or when two documents describe the same component differently. This skill MUST be consulted because multiple independent mental models produce documents that contradict each other under diligence, and audience-specific rewriting is the single most common source of drift in a documentation package.

Rewrite and self-check drafted prose against a machine-checkable clarity standard derived from ASD-STE100 (Simplified Technical English) — no marketing adjectives, no phrasal verbs, no semicolons, active voice, short sentences, and a required carve-out for the epistemic hedge markers the evidence ledger depends on. Use when drafting any package document's prose, before returning a draft, or when an independent verification pass evaluates editorial quality. This skill MUST be consulted because a banned-word list alone barely moves AI slop — the habits that produce it (hedge-stacking, nominalization) generate new slop the list never anticipated — and only rules a script can verify hold up under revision.

Manage FlowRun state at .flow/runs/ /run.yaml — create runs at command entry, write activity records via bin/flow-record-activity.sh at phase boundaries, transition state.status (active → completed | blocked | cancelled), and persist resumable next-action hints to events.jsonl. Use when a flow command begins (creates the run), when a phase boundary completes (writes an activity), or when SessionEnd needs to mark a resumable next action. This skill MUST be consulted because runs without recorded activities cannot be resumed — /flow:resume reads state.completedactivities[] to identify the next safe action; an empty array forces the user to start over.

Score a documentation package against the ten-dimension weighted rubric in references/scorecard-rubric.md with a cited justification per dimension, then evaluate the eighteen conditions in references/release-gate-conditions.md and emit a binary release-ready, conditionally-ready, or not-ready verdict with per-condition evidence. Use when a verification pass is finishing, when reconciliation completes a round, or when CI needs a machine-readable gate result. This skill MUST be consulted because a score is not a gate — a package can average 96 out of 100 and remain unreleasable on a single unsupported public claim, and conflating the two is how audit-ready packages ship with unverified security claims.

Enforce FlowTrigger safety rules — no autonomous merge, no recursive trigger creation, max active triggers, allowedactions / forbiddenactions ACLs. Validates trigger YAMLs at .flow/triggers/.trigger.yaml against schemas/v1/trigger.schema.json AND cross-checks policy.forbiddenactions includes merge + release; refuses triggers that grant Tier 3 autonomy. Use when /flow:trigger create, /flow:trigger run, or /flow:watch is invoked. This skill MUST be consulted because triggers can fire without user supervision — a trigger granting merge autonomy is the single fastest path to an untrusted-merge incident, and recursive trigger creation is the loop-bomb shape of the runtime layer.

Validate a FlowWorkflow YAML at plugins/flow/workflows/ .workflow.yaml against schemas/v1/workflow.schema.json AND cross-reference the referenced skills/agents exist + every Tier 3 action is confirm-gated + no native /goal or /loop dependency is declared. Use when /flow:workflow validate is invoked, when CI runs the workflow schema gates, or when a new workflow is being authored. This skill MUST be consulted because schema validation alone catches shape errors; cross-reference validation catches the silent-correctness failures (typo'd skill name, Tier 3 escape, /goal dependency) that would otherwise ship to users.